DEV Community

Nat
Nat

Posted on Originally published at aidenai.io

Browser Agents Can Use the Web. Aiden Is Built to Use the Device.

Perplexity Comet is capable inside a browser session. Aiden is being built as a device automation agent for work that moves across visible phone and computer interfaces — beyond what any browser session can reach.

Comet deserves the comparison

Perplexity positions Comet as an AI-first browser and personal assistant, available across Mac, Windows, iPhone, iPad, and Android, with a usable free tier. For research, page-level questions, tab-heavy synthesis, and supported web actions, that browser-native model is genuinely strong.

The distinction isn't that browser agents are less useful — it's that every agent works through a control surface. Comet's surface is the browser session: tabs, webpages, browser-accessible documents, web apps. Aiden's surface is the connected device itself.

This is a real, expanding category, not a single-product comparison:

  • Fellou — explicit "self-driving browser" positioning for multi-step web tasks
  • Dia (The Browser Company) — strong cross-tab context, though more suggestive than autonomous
  • Opera Neon — experimental, agentic, broader creation-workflow emphasis

Each reflects the same direction: AI moving from answering questions to participating in interface work. The practical difference is where that participation can continue.

Where the browser boundary actually shows up

Workflow moment Browser-session strength Where the boundary appears
Researching sources Web search, reading, tab comparison Task stays browser-contained
Filling a web form Browser-rendered fields, on-page actions External identity checks or system dialogs
Saving findings Web notes, browser-accessible docs Native notes/desktop app needs another path
Working with local files Downloads, web uploads File pickers, desktop software, permission prompts
Continuing on a phone Mobile browser context Native apps, device settings, cross-app handoffs
flowchart LR
    A[User instruction] --> B{Current control surface}
    B --> C[Browser session]
    C --> D["Tabs, webpages, web apps"]
    D --> E[Browser-supported action]
    E --> F[Review browser outcome]
    B --> G[Connected device interface]
    G --> H["Browser, native app, desktop app, or system dialog"]
    H --> I[Observe visible state]
    I --> J[Propose next action]
    J --> K[User confirms, redirects, or stops]
    K --> L[Execute supported interaction]
    L --> M[Review outcome]
Enter fullscreen mode Exit fullscreen mode

How Aiden actually works

Aiden's firmware repo documents a development-board implementation: HDMI-based screen capture, USB HID input. It observes a connected screen and sends keyboard, pointer, or touch-style input — no app-specific integration required on the target device.

That's evidence of a documented dev-board implementation. It's not a claim of universal compatibility, task-success guarantees, or consumer availability. Aiden's being developed for Android and iPhone workflows specifically, and the team is upfront that real-device interaction has real constraints — setup conditions, OS versions, accessibility settings, interface variability.

Choosing between the two isn't a category war

Need Browser-first (Comet) Device-level (Aiden)
Research across webpages Strong fit Relevant only if research continues onto a device
Native mobile app handoff Depends on integrations Core workflow category Aiden targets
Desktop application interaction Depends on external access paths Relevant when visible desktop interfaces matter
Consequential decisions User review required User confirmation/interruption stays central

Aiden isn't presented as a Comet replacement. Use a browser agent for web research and browser-native tasks; use the device-level approach when a real interface handoff becomes necessary.

Repo: github.com/AidenAI-IO/aiden-firmware
Discord: discord.com/invite/bcJavjcnYz — engineers active, technical questions welcome.

Top comments (0)